Does artificial intelligence redefine nuclear-to-cytoplasmic ratio threshold for diagnosing high-grade urothelial...

The Paris System (TPS) introduced standardized nuclear-to-cytoplasmic (N/C) ratio thresholds for urine cytology to improve high-grade urothelial carcinoma (HGUC) detection, but these criteria remain subjective. This study used AIxURO, an artificial intelligence-based model, to measure N/C ratio and nuclear area to identify abnormal cells in whole slide images (WSIs).

A total of 106 urine cytology slides from 46 lower urinary tract (LUT) and 60 upper urinary tract (UUT) cancer cases, diagnosed as atypical urothelial cell (15.1%), suspicious for high-grade urothelial carcinoma (SHGUC) (23.6%), and HGUC (61.3%), with biopsy-confirmed HGUC or carcinoma in situ (CIS), were digitized and analyzed by AIxURO. The model quantified suspicious and atypical cells, N/C ratios, and nuclear areas, with statistical differences assessed using Kruskal-Wallis tests.

AIxURO identified fewer suspicious cells than atypical cells (20.5 vs. 242.0, p < .001). Suspicious cells had higher N/C ratios (0.66 vs. 0.58, p < .001) and larger nuclear areas (102.3 vs. 85.7 µm², p < .001). Although N/C ratios did not differ significantly between UUT and LUT cases, nuclear areas varied among abnormal cells (CIS: 101.5 µm²; HGUC: 83.5 µm²). In HGUC cytology cases, the CIS category had larger nuclear areas than HGUC for both suspicious (116.3 vs. 100.4 µm²) and atypical cells (101.5 vs. 82.2 µm²).

AIxURO provides objective quantification of N/C ratios and nuclear areas, refining TPS criteria for distinguishing suspicious from atypical cells. A lower N/C ratio cutoff (0.66) for SHGUC/HGUC may be more appropriate than the TPS threshold (>0.7). Findings support using consistent N/C ratio criteria across UUT and LUT cases.